3 Advances in Geologic Mapping, Databases, and Dissemination: Establishing fundamental resources for the earth science community

This session will highlight new geologic maps, innovations in mapping techniques, and related advances in data management, web accessibility, and applications. We welcome submissions with relevance to hazard assessments, natural resource evaluations, and water-land management.

Monday, 9 March 2026
9:00 AM - 12:00 PM
RCC, 105

Expanding Karst and Sinkhole Hazard Assessment Using Deep Learning Image Segmentation

Paper number: 3-2
Program: Technical Sessions
Time: 9:25 AM – 9:45 AM

Authors

Abstract

Kentucky is well-known for its prominent karst landscape characterized by rolling hills, caves, sinkholes, and springs. Karst developments, especially sinkholes, can cause damage to buildings, highways, and other infrastructure. Understanding karst-related hazard and its associated risk is important for urban planning, land management, and hazard mitigation for many Kentucky communities. To enhance karst and sinkhole hazard evaluation for Kentucky, we have developed a hazard assessment method that relies on accurate lidar-derived sinkhole maps and karst characteristics of bedrock geology. Accurately mapping sinkholes from lidar, however, is a slow process that requires manual visual inspection and field checking. Currently, lidar-derived sinkhole maps cover about one-third of Kentucky’s counties. As a result, the karst hazard assessment method cannot be expanded to all karst areas in Kentucky. Image segmentation is a deep learning computer vision process that identifies objects and their boundaries directly from images and thus can be applied to automatically map sinkholes. To determine whether image segmentation could allow our karst hazard evaluation process to be expanded to areas without lidar-derived sinkhole maps, we first trained a U-net image segmentation model using high-resolution lidar-derived elevation data and inventories of manually mapped sinkholes. We then applied the trained U-net model to automatically map sinkholes in Grayson County, Kentucky, and used the resulting maps in our karst hazard assessment for the county. The hazard assessment incorporating image segmentation yielded very similar results to the assessment obtained using manually lidar-derived sinkholes, providing a viable and rapid way to expand our karst hazard method to other areas. In addition, as multitemporal lidar data are increasingly available for many regions, deep learning image segmentation also offers a timely way to update karst and sinkhole hazard assessments to reflect the sinkhole information captured in the latest lidar elevation data.

Geological Society of America Abstracts with Programs. Vol. 58, No. 1

doi: 10.1130/abs/2026SE-411589

© Copyright 2026 The Geological Society of America (GSA), all rights reserved. Permission is hereby granted to the author(s) of this abstract to reproduce and distribute it freely, for noncommercial purposes. Permission is hereby granted to any individual scientist to download a single copy of this electronic file and reproduce up to 20 paper copies for noncommercial purposes advancing science and education, including classroom use, providing all reproductions include the complete content shown here, including the author information. All other forms of reproduction and/or transmittal are prohibited without written permission from GSA Copyright Permissions.